Long-term application of green manure (GM) and nitrogen (N) fertilizers markedly improved soil fertility and boosted rice yield in ecologically fragile karst paddy fields. However, the precise response mechanisms of the soil bacterial community to varying amounts of green manure alone and in combination with N fertilizer in such environments remain poorly elucidated. In this study, we investigated the soil bacterial communities, keystone taxa, and their relationship with soil environmental variables across eight fertilization treatments. These treatments included group without N addition (N₀M₀, no N fertilizer and no GM; N₀M_22.5, 22.5 t/ha GM; N₀M₄₅, 45 t/ha GM, N₀M_67.5, 67.5 t/ha GM) and group with N addition (NM₀, N fertilizer and no GM; NM_22.5, N fertilizer and 22.5 t/ha GM; NM₄₅, N fertilizer and 45 t/ha GM; NM_67.5, N fertilizer and 67.5 t/ha GM). The results revealed that increasing green manure input significantly boosted rice yield by 15.51–22.08% and 21.84–35% in both the group without and with N addition, respectively, compared to N₀M₀ treatment. Moreover, with escalating green manure input, soil TN, AN, AK, and AP showed an increasing trend in the group without N addition. However, following the addition of N fertilizer, TN and AN content initially rose, followed by a decline due to the enhanced nutrient availability for rice. Furthermore, the application of a large amount of N fertilizer decreased the C: N ratio in the soil, resulting in significant changes in both the soil microbial community and its function. Particularly noteworthy was the transition of keystone taxa from their original roles as N-fixing and carbon-degrading groups (oligotrophs) to roles in carbon degradation (copiotrophs), nitrification, and denitrification. This shift in soil community and function might serve as a primary factor contributing to enhanced nutrient utilization efficiency in rice, thus significantly promoting rice yield.

Excessive nitrogen (N) fertilization in rice systems has caused soil degradation and reduced N use efficiency. Green manure, especially Astragalus sinicus (Chinese milk vetch), provides a sustainable alternative, but the microbial and functional gene mechanisms underlying its interaction with reduced N input remain unclear. In this study, a field experiment was conducted at Dingdian Village, Natong Town, Long’an County, Nanning City, Guangxi Province, China (107°51′21″ E, 23°00′41″ N) during the 2018–2019 rice growing seasons. Four treatments were established: conventional N fertilization (N100), 20% N reduction (N80), green manure plus full N (GMN100), and green manure plus 20% N reduction (GMN80). Soil physicochemical traits, microbial community composition, and carbon-cycling functional genes were analyzed using high-throughput sequencing and metagenomic profiling. Compared with N100, GMN80 significantly increased soil organic matter (by 21.3%), microbial biomass carbon (by 32.6%), and available phosphorus (by 17.8%). The Shannon index rose from 4.18 to 4.63, while Proteobacteria and Actinobacteria increased by 9.5% and 7.2%, respectively. Functional genes encoding glycoside hydrolases (GH5, GH9) and carbohydrate esterases (CE1, CE10) were enriched by 25–40%, with upregulation of carbon fixation (rbcL) and methane metabolism (mcrA) genes. Integrating A. sinicus with moderate N reduction improves soil fertility, stimulates microbial diversity, and enhances carbon turnover efficiency, offering a practical pathway toward sustainable low-carbon rice production.

In karst agroecosystems, the stability of soil organic carbon (SOC), which is essential for sustainable land management, is strongly influenced by the carbonate parent material. However, how land use and lithology interactively affect the vertical distribution and stabilization of SOC fractions remains unknown. This study investigated the vertical distribution (0–70 cm) of SOC fractions (particulate, POC; mineral–associated, MOC; recalcitrant, ROC; microbial biomass, MBC; dissolved, DOC; permanganate–oxidizable, LOC) in paddy and upland systems overlying limestone and dolomite. Results showed that land use and lithology jointly determined depth–dependent patterns of SOC fractions. The key finding was a depth–dependent reversal of land use effects: paddy soils had higher concentrations of SOC fractions (especially MBC and LOC) in the top 0–20 cm, whereas upland soils had higher concentrations of DOC, ROC, and POC below 50 cm. This pattern underscores an overlooked deep carbon sequestration potential in aerated uplands. Furthermore, lithology exerted fundamental control on SOC stability, with limestone soils exhibiting higher MOC/SOC and MOC/POC ratios than dolomite soils, indicating stronger organo–mineral association. Random Forest and Structural Equation Modeling revealed that soil aggregate stability is a key factor influencing the stability of SOC. These findings demonstrate that effective SOC management in karst agroecosystems must consider both surface land use decisions and deep lithological constraints. Specifically, promoting aggregate stability offers a more targeted pathway for long–term carbon stabilization in these ecologically fragile agricultural systems.